Memories are devices able to store and to make accessible information stored in binary form as bits, and they can be subdivided in various categories according to the storage capacity and to the time necessary to retrieve the information stored therein.
Semiconductor memories are made in MOS (metal-oxide-semiconductor) technology on a semiconductor material substrate, typically single crystalline silicon, and are called non-volatile when they retain the information for considerable times and in absence of power supply.
Among non-volatile MOS memories, a particularly important class is that constituted by floating gate devices, in which the single cell is constituted by a MOSFET (metal-oxide-semiconductor field effect transistor) in which the gate electrode includes an additional electrode (floating gate) placed between the channel and the control electrode (control gate), completely surrounded by electrical insulation and separated by a dielectric from the control gate.
The information stored in the cell is represented by the charging state of the gate electrode, that is modified by either injecting electrons from the FET into the floating gate (writing) or removing them (erasing).
In absence of relatively high voltages applied to the FET electrodes, the floating gate charge remains almost unaltered in time because the electrons remain in such electrode without dispersing into the surrounding environment, thanks to the presence of insulating layers that surround the floating gate.
Among the floating gate non-volatile MOS memories, a dominant position is occupied by flash memories, whose main features are given by the possibility of being written and erased electrically, by random access not only for reading but even for writing, and by the considerable high integration density, due to the presence of a particularly compact single transistor elementary cell.
In flash memories, the mechanism adopted for writing, or programming, a memory cell is the injection into the floating gate of “hot” electrons coming from the channel and “heated” by the application of a suitable potential difference between source and drain.
In a write operation, some ten of thousands of electrons are injected into the floating gate, and the retention thereof inside the floating gate provides the memory non-volatility.
The physical mechanism adopted for erasing a flash memory cell, an operation in which the floating gate substantially is emptied of the electrons injected during the writing, is the tunneling through a dielectric of the electrons from the floating gate to the source, made possible by the application of a suitable potential difference between the control gate and the source.
The efficiency of this charge transfer process is measured by the capacitive coupling between control gate and floating gate, that is expressed in terms of a capacitive coupling coefficient αG, defined as the ratio between the capacitance CCG of the control gate and the total capacitance CTOT associated with the gate electrode, that also takes into account the capacitive couplings due to the presence of the source Cs, of the drain CD, of the tunnel oxide and of the channel CFG:αG=CCG/CTOT=CCG/(CCG+CFG+CS+CD).
According to a model widely in use, the sequence of control gate/dielectric/floating gate layers can be represented as a capacitor with plane and parallel plates separated by a dielectric, the floating gate and the control gate being the two plates thereof.
The capacitive coupling coefficient αG is thus proportional to the capacitance CCG=∈diel (AFG/Tdiel) where ∈diel is the dielectric constant of the dielectric, AFG is the area of the floating gate surface facing the control gate, and Tdiel is the dielectric thickness.
Consequently, the capacitive coupling coefficient αG strongly depends on the shape and the size of the floating gate and, in particular, it is proportional to the floating gate surface area facing the control gate.
In FIGS. 1 and 2 there are shown, in simplified way, respectively a cross section and a circuit scheme of a portion of a matrix 200 of flash memory cells, such cross section being taken along a direction AA′ and along a direction BB′ perpendicular thereto.
FIG. 1 evidences the layered structure of the gate electrode region 4, the wells of the source region 1 and drain region 2, and the central electrically active region 10, formed by the FET channel within a silicon single crystalline substrate 3.
Over the FET channel 10 the gate-electrode region 4 is constituted by:                a first thin dielectric layer 5, typically silicon oxide, called tunnel oxide;        the floating gate 6, usually formed by a heavily doped polycrystalline silicon layer;        a second dielectric layer 7, made for example of a succession of SiO2/Si3N4/SiO2 thin layers called ONO (acronym for oxide-nitride-oxide), that covers the floating gate 6;        the control gate 8, typically formed by a heavily doped polycrystalline silicon layer.Along the direction BB′, the memory cells 100 are separated by insulating regions 9, that in the currently more advanced technologies are of STI (shallow trench isolation) type, i.e., they are constituted by trenches in the single crystalline silicon substrate 3, filled up by one or more dielectric layers.        
Typically, the memory cells 100 are organized in a matrix structure, that, as shown by way of example in FIG. 2 for a NOR matrix 200 of flash type memory cells 100, is arranged in rows 21, called word lines, running along the direction BB′, and columns 22, called bit lines, running along the direction AA′.
In the matrix 200, the control gates 8 of the memory cells 100 form the word lines 21, and along this direction source connection lines 24, constituted by semiconductor material, extend in regular intervals, for example every sixteen cells, running parallel to the bit lines 22.
The bit lines 22, which constitutes the drain connection lines, are formed by conductor material, typically a metal or an alloy of one or more metals (for example Al, AlCu, Cu, W . . . ) and run perpendicularly to the word lines 21.
In the direction of the bit lines 22, the drains 2 of adjacent memory cells 100 face each other, and in correspondence to each pair of faced drains 2, drain electric contacts 23 are provided along the bit lines 22, connecting the drains 2 to the bit lines 22.
Also the sources 1 of adjacent memory cells 100 face each other in the direction of the bit lines 22, and the diffusion source lines 26 connect them to source connection lines 24 by means of source electric contacts 25, along the direction of the word lines 21.
A conventional process for fabrication of flash memories calls for the formation of the isolation regions 9, for example of STI type, and of the floating gates 6 of the memory cells 100 through the following phases:
1. On the single crystalline silicon substrate 3, a sufficiently thin dielectric layer is grown, of thickness ranging from 10 to 20 nm, called pad oxide;
2. On the pad oxide, a silicon nitride layer of thickness typically ranging from 100 nm to 200 nm is deposited, that has the function of stop layer for the following planarization treatments;
3. The areas where the STI type isolation regions 9 will be made are defined by lithography;
4. The nitride layer and the pad oxide are removed in sequence from these areas, and trenches of the desired depth, typically about 150 nm, are formed inside the single crystalline silicon bulk 3;
5. The trenches are filled with one or more layers of dielectric material, that as a whole are called field oxide;
6. The field oxide is planarized, typically using the CMP (chemical mechanical polishing) technique, in such a way as its exposed surface is flush with that of the still present nitride portion;
7. The exposed field oxide surface level is lowered, typically by a wet etch in hydrofluoric acid (HF), so that at the end of the process the height difference between the field oxide surface and the surface of the single crystalline silicon substrate 3 is not too high (−20 nm).
After having formed in this way the STI type isolation regions 9, the process of formation of the floating gates 6 of the memory cells 100 proceeds with the following phases:
8. The portion of nitride still present is removed;
9. A sacrificial dielectric layer called sarox is grown, of thickness approximately equal to 10 nm;
10. Through the sarox layer, some phases of dopant implantation are performed, necessary to the operation of the memory cell 100;
11. The sarox layer and the pad oxide layer are removed by wet etching;
12. The tunnel oxide 5 is grown, of thickness ranging from 5 nm to 10 nm;
13. A polycrystalline silicon layer is deposited, of about 100 nm thickness;
14. The polycrystalline silicon layer is defined by lithography and etching, so as to form the floating gate 6 of the memory cell 100;
15. The layer of ONO (oxide-nitride-oxide) 7 is deposited, covering the floating gate 6;
16. The control gate 8 is formed, typically made of heavily doped polycrystalline silicon.
In the last generation technologies, the memory-cell dimensions 100 are so small that to use traditional lithography for the definition of the floating gate 6, mask alignment is often critical. In fact, a possible misalignment, even minimum, in the lithographical definition phase (14 above) of the polycrystalline silicon layer, might be too great for such small geometries, and cause the memory cell 100 to be incorrectly defined.
On the other hand, the quest to reduce the memory dimensions becomes more and more pressing as the technology progresses, and this demand translates into the necessity of devising technological solutions and innovative integrated structures that allow combining, in such devices, an optimal electric behavior with minimal geometric dimensions.
Therefore, the possibility has been investigated of making a memory cell with self-aligned insulation, so called because the electrically active regions and the floating gate are defined simultaneously. The floating gate is thus automatically aligned to the isolation regions. An example of flash memory cell 300 with self-aligned insulation is shown in FIG. 3 in schematic transversal section along the direction BB′, not to scale and limited to the layers of interest. Along the direction AA′, the flash memory cell 300 has a structure similar to the memory cell 100 shown in FIG. 1.
In detail, the solution with self-aligned insulation provides for forming isolation regions 9, for example of STI type, and the floating gates 6A of the memory cells 300 through the following phases:
1A. The tunnel oxide 5 is grown on the single crystalline substrate 3;
2A. A polycrystalline silicon layer, forming the floating gate 6A, of thickness about of 200 nm is deposited;
3A. A silicon nitride layer of thickness typically ranging from 100 nm to 200 nm, having the function of stop layer for the following planarization treatments, is deposited;
4A. The STI type isolation regions are defined by lithography and etch of the sequence of silicon nitride layer, the polycrystalline silicon layer, the tunnel oxide 5 layer and single crystalline substrate 3, up to formation of trenches of desired depth, typically about of 150 nm, inside the single crystalline silicon substrate 3.
The floating gates 6A thus defined are self-aligned to the type STI isolation regions.
5A. The trenches are filled with the field oxide;
6A. The field oxide is planarized, typically using the CMP (chemical mechanical polishing) technique, in such a way that the exposed surface thereof is flush with that of the silicon nitride layer.
7A. The silicon nitride layer is removed by isotropic etching;
8A. The level of the exposed surface of the field oxide is lowered, typically by wet etching in hydrofluoric acid (HF), in such a way to uncover the side walls of the floating gate 6A.
9A. The ONO (oxide-nitride-oxide) layer 7, covering the floating gate 6, is deposited;
10A. The control gate 8, typically made of heavily doped polycrystalline silicon, is formed.
Since in this way the floating gate 6A is self aligned to the STI-type isolation regions 9, this process flow allows obtaining memory cells 300 of reduced dimensions compared to the memory cells 100 made with the traditional process flow.
Regretfully, for a given thickness of deposited polycrystalline silicon, this last-generation process flow produces floating gates 6A of such a morphology that strongly reduces the capacitive coupling coefficient αG between the floating gate 6A and the control gate 8.
In fact, comparing the morphology of the traditional floating gate 6 of the memory cell 100 with that of the floating gate 6A of the memory cell 300 with self-aligned insulation as represented in FIG. 4, it can be noticed that the traditional floating gate 6 has some protrusions, so-called “wings”, that extend over the adjacent STI-type isolation regions 9. Such wings are instead missing in the case of the floating gate 6A of the memory cell 300 with self-aligned insulation.
The lack of these “wings”, due to the complexity of the process phases that are used for the realization of the floating gate 6A, causes the width W2 of the floating gate 6A to be smaller than the width W1 of the traditional floating gate 6.
Besides, the effective thickness T2 of the floating gate 6A is also lower than that of the traditional floating gate 6, indicated with T1.
This difference is due to the fact that in the process flow realizing the floating gate 6A the lowering step of the level of the exposed field oxide surface (step 8A) is performed after the growth phase (step 1A) of the tunnel oxide 5. Therefore a margin, on the order of the about ten nanometers, is required in the lowering phase (step 8A) of the exposed field-oxide surface so as to preserve the tunnel oxide 5.
Accordingly, the side walls of the floating gate 6A are partially covered by field oxide, and therefore the effective thickness T2 of the floating gate 6A is lower than the effective thickness T1 of the traditional floating gate 6.
These differences in the geometric dimensions of the floating gate 6 and 6A are reflected in the values of the capacitive coupling coefficients αG of the traditional memory cells 100 and 300 with self-aligned insulation, whose ratio is expressed by the relationship:αG(300)/αG(100)÷(W2+2T2)/(W1+2T1)from which it can be deduced that, W2 and T2 being lower than W1 and T1, the capacitive coupling coefficient αG of the memory cell 300 with self-aligned insulation is lower than the corresponding capacitive coupling coefficient αG of the traditional memory cell 100.
It has been verified experimentally that this difference can be estimated to be of order of at least 10 percent.
Such an efficiency loss in the capacitive coupling αG leads to an increase in the erasing time of the memory cell 300 with self-aligned insulation, since the charge transfer process from the floating gate 6A to the source region 1 is less efficient, and thus may cause the final erase voltage to be higher than with the floating gate 6.
The management of high erase voltages is often burdensome and can cause phenomena such as degradation of the quality of the active oxides, worsening the characteristics of reliability of the memory cell 300 with self-aligned insulation, the electric performance of which may be accordingly limited.
The formation of the floating gate 6A starting from a thicker deposited polycrystalline silicon allows partial recovery of the loss of capacitive coupling between floating gate and control gate, but this may introduce other problems; in particular, the filling by field oxide of the isolation trenches may become difficult because of the increased depth of the trench to be filled.